TBM Performance Prediction Using Statistical and Machine Learning Algorithms: Case Study of Slurry TBM Used in Mumbai Metro Rail Tunnel Line–3 (UGC-03), India

ABSTRACT Estimating the performance of a tunnel boring machine (TBM) is vital for enhancing mining productivity and reducing tunnelling operations-related risks. Penetration rate (PR) is the most prominent TBM performance indicator, which majorly influences the project completion time, contractual activities, and overall cost of the project. Prediction of PR considering varying geotechnical parameters needs to be addressed. This study comprises 235 data points collected from a 3.572 km long twin-tunnel bored using two refurbished articulated single-shield slurry TBM, employed in a metro project in package-03, (UGC- 03) of Mumbai metro rail corporation line-3 (MMRCL-3), in Mumbai, India. The PR of the TBM is predicted using; uniaxial compressive strength (UCS), point load test (PLT (I50)), rock mass rating (RMR), rock quality designation (RQD), porosity (n), and Brazilian tensile strength (BTS) as the input parameters considering two statistical approaches i.e., multivariate linear regression (MVLR) and multivariate non-linear regression (MVNLR) and five machine learning (ML) algorithms i.e., random forests (RF), K-nearest neighbour (KNN), XG-Boost, support vector regression (SVR) and artificial neural networks (ANN). A 90: 10 train: test split was used on the 235 data points. The model developed using XG-Boost regression predicted the most accurate PR with R2 = 0.7877 / RMSE = 0.2113, where RF was the second-best model and closest to XG-Boost with R2 = 0.7679 / RMSE = 0.2200. The developed models indicate a considerable relationship between the input parameters and PR, and PR can be predicted accurately.